US 7155319 B2
A method for controlling liquid delivery in a processing chamber. The method includes generating an analog input (AI) signal proportional to a process variable and calculating an analog output (AO) signal based on a setpoint and a deadband. The setpoint is a target value of the process variable and the deadband is an allowable tolerance around the setpoint that determines when the control logic is activated to control the process variable. The method further includes transmitting the AO signal to a control device and adjusting the process variable proportional to the value of the AO signal.
1. A method for controlling liquid delivery in a processing chamber, comprising:
generating an analog input (AI) signal proportional to a process variable;
calculating an analog output (AO) signal based on a setpoint and a deadband, wherein the setpoint is a target value of the process variable and the deadband is an allowable tolerance around the setpoint that determines when the control logic is activated to control the process variable;
transmitting the AO signal to a control device; and
adjusting the process variable proportional to the value of the AO signal.
2. The method of
3. The method of
4. The method of
5. The method of
if (setpoint−deadband)<flow<(setpoint+deadband), then AO=AOold, wherein flow is the value of the process variable based on the most recent AI signal;
if flow>(setpoint+deadband), then AO=AOold−(correction×correctscale), wherein correction is a multiplier that controls the rate at which the process variable is adjusted to within the deadband and wherein correctscale is a restriction coefficient that controls the rate at which the process variable is adjusted to within the deadband; and
if flow<(setpoint−deadband), then AO=AOold+(correction×correctscale).
6. The method of
7. The method of
incorporating a time delay from about 0.0 seconds to about 5.0 seconds into the control logic to prevent the control device from being adjusted for the duration of the time delay; and
fixing the AO for the duration of the time delay.
8. The method of
9. The method of
10. The method of
11. The method of
12. The method of
13. The method of
14. The method of
15. The method of
16. The method of
17. The method of
18. The method of
19. The method of
20. The method of
21. The method of
22. The method of
if (setpoint−deadband)<flow<(setpoint+deadband), then AO=AOold;
if flow>(setpoint+deadband), then AO=AOold−(correction×correctscale); and
if flow<(setpoint−deadband), then AO=AOold+(upcorrection×correction×correctscale), wherein upcorrection is a multiplier constant that controls the rate at which the process variable is adjusted during a ramp up to the setpoint.
23. The method of
24. A method of monitoring drift of a closed loop control system, comprising:
generating an analog input signal proportional to a process variable being controlled to a target setpoint;
calculating an analog output signal based on the analog input signal and the target setpoint for the process variable;
determining a drift has occurred if the analog output signal is one of less than a lower voltage warning or greater than an upper voltage warning; and
generating a warning.
25. The method of
26. The method of
27. The method of
1. Field of the Invention
Embodiments of the invention generally relate to the closed loop control of a process variable and more specifically to closed loop control of liquid delivery for wet processing of substrates in the fabrication of electronic devices.
2. Description of the Related Art
Wet processing of substrates in the fabrication of electronic devices, particularly in electrochemical plating (ECP) and chemical/mechanical polish (CMP) applications, often require precisely controlled chemical deliveries for very short process steps.
Typically, when a substrate is treated in a process chamber, a series of treatments will take place sequentially in the same chamber. This series of treatments, or process steps, is generally referred to as a process recipe. Process recipes are application specific and, therefore, vary depending on which electronic device manufacturer is operating the chamber, what device is being fabricated on the substrate, and sometimes which particular film of the device is currently on the surface of the substrate (i.e., metal 1 vs. metal 2, etc.). In the case of the chemical delivery steps in a wet wafer processing chamber, the process steps in a process recipe are often as short as 10–15 seconds. Under these conditions, accurate and repeatable control of liquid chemical delivery is difficult, and failure to provide such control directly affects substrate quality and device yield. Standard control methods for this application include open loop control and closed loop control.
A simple method of controlling liquid delivery with open loop control uses a fixed orifice with constant pressure. This method controls flow with an adjustable, fixed orifice, for example a needle valve, and a pressure regulator, which maintains a constant fluid pressure upstream of the needle valve and therefore maintains a constant flow rate. For a number of reasons such control has proven inadequate for chemical delivery to a wet processing chamber for the processing of a semiconductor substrate. Chronic problems with this method include: drift from setpoint of both the fixed orifice and the pressure regulator, poor repeatability chamber-to-chamber and substrate-to-substrate, cross-talk between chambers (i.e., actual flow rate to one chamber is affected by whether there is flow to other chambers at the same time), difficulty in monitoring flows without introducing additional sensors, and an inability to incorporate alarms when the flow rate is too far from the target setpoint. Additionally, any changes to the liquid delivery system that affect liquid flow cannot be compensated for with open loop control. Examples of changes include: liquid tubing re-routed or kinked, liquid tubing replaced with tubing of different length or inner diameter, liquid delivery nozzles changed, and flow regulator performance altered due to mechanical wear. When such changes occur, re-calibration is required. Therefore, in order to achieve stable and repeatable DI water and chemical flow to process chambers it is necessary to have a closed loop flow control system.
Closed loop control is based on continuously modifying the desired process variable based on the measured value of the process variable and the setpoint for the process variable. This involves measuring the desired process variable with an appropriate sensor, generating a signal proportional to the measurement, sending this signal to a computer, processing the signal in the computer (using a control algorithm that determines what adjustment needs to be made to the process variable in order for the process variable to be closer to the target setpoint value), and outputting a signal to a control device that is proportional to the desired correction. The amount of correction is a function not only of how much the process variable needs to be adjusted, but also other factors: sampling rate, response times of the control device and sensor, and physical factors such as valve sizes and sensor locations. The process variable is then measured again and the process is repeated, typically at a relatively high frequency, for example every 50 milliseconds. Closed loop control allows for stable and repeatable control of a process variable. The state-of-the-art method used for closed loop control is PID (proportional, integral, derivative) control. With the correct tuning of PID control parameters, this algorithm can provide process variable control over a wide range of situations.
In the case of liquid delivery in a wet substrate processing chamber, a typical closed loop control system might consist of an ultrasonic flow meter or vortex flow meter (FM) as a flow sensor, a pneumatically controlled pressure regulator as a flow regulator, a dedicated computer processing the input signal from the FM with a PID algorithm, and a voltage-to-pressure transducer to convert the output signal voltage from the computer to a proportional pneumatic pressure to operate the control device. For situations involving continuous control of the flow over relatively long periods of time, for example minutes or hours, this system can work well. However, for short process recipe steps that require controlling the flow from full off to a given setpoint in a short time, PID control is not very effective. This is because when tuning a PID control loop, there is a direct trade off between quick response time and minimal overshoot past the setpoint.
In the case of liquid chemical delivery for the treatment of semiconductor substrates, it is very important to avoid overshoot of the flow rate setpoint. This is because overshoot can result in splashing of chemical in the process chamber, which can cause serious defects on the surface of the substrate. Also, for some semiconductor applications, the liquid chemicals must be delivered at precise flow rates, otherwise, serious defects can occur.
For a sudden change in setpoint that only lasts a short time, PID control will still be stabilizing the flow (i.e., hunting above and below the new setpoint) during some or the entire process recipe step. This results in chronic and often serious variations in the process each time the recipe is run. Another problem with PID control in some situations is the fact that it is a calculation intensive method for determining how much the process variable needs to be corrected. This is because no matter how close the process variable is to the target setpoint, the PID algorithm will continue to calculate a correcting adjustment. Often this is not important, but if the computer performing these calculations is shared by a large number of sensors and control devices, for example as with the system controller for a wafer processing system, the response time of the entire system can be slowed, dramatically affecting the ability of the wafer processing system to function. Additionally, if the control device in such a control loop is subject to a significant amount of hysteresis, more sophisticated algorithms need to be incorporated into the PID calculation to offset this effect. This makes the control process even more calculation intensive.
Another issue encountered when controlling fluid flow or other process variables is drift. Drift is caused by long-term systemic changes that increasingly affect the control of a process variable and/or the calibration of sensors or control devices. Factors that contribute to the drift of sensors, control devices, and other mechanical components include normal wear and tear, accumulated contamination, and loss of calibration. Drift can be difficult to detect since it generally takes place over a long period of time (weeks or months) and a conventional closed loop control will compensate until it can no longer keep the process variable at setpoint. An example of this is the mass flow controller (MFC) used in semiconductor manufacturing applications, which generally operates as a self-contained PID closed loop controller that supplies process gas at a specific flow rate. Typically, a central system controller will communicate a desired setpoint to the MFC for a particular recipe and the MFC's internal closed loop control will maintain gas flow at that setpoint. If drift of the gas delivery system being controlled occurs over time (for example due to a loaded filter), the MFC will continue to compensate to keep the process variable at setpoint. It will only send an alarm to a central system controller once it is fully open and still cannot control the gas flow to the setpoint. For most semiconductor processes, this is too late to avoid compromising substrate quality/yield. To prevent this, conservative preventive maintenance (PM) schedules are typically followed for semiconductor manufacturing equipment, for example early filter replacement. This leads to greater system downtime and inefficient use of spare parts and labor.
Therefore, there is a need for a method of closed loop control that:
Various embodiments of the invention are directed to a method for controlling liquid delivery in a processing chamber. The method includes generating an analog input (AI) signal proportional to a process variable and calculating an analog output (AO) signal based on a setpoint and a deadband. The setpoint is a target value of the process variable and the deadband is an allowable tolerance around the setpoint that determines when the control logic is activated to control the process variable. The method further includes transmitting the AO signal to a control device and adjusting the process variable proportional to the value of the AO signal.
Various embodiments of the invention are also directed to a method of monitoring drift of a closed loop control system. The method includes generating an analog input signal proportional to a process variable being controlled to a target setpoint; calculating an analog output signal based on the analog input signal and the target setpoint for the process variable; determining a drift has occurred if the analog output signal is one of less than a lower voltage warning or greater than an upper voltage warning; and generating a warning.
Various embodiments of the invention provide improved process control over the prior art in applications where it is critical to have a quick response time, fast stabilization at the setpoint, and minimal overshoot. When a sudden change in setpoint of a process variable occurs, a control device is initially opened or closed to a best guess position based on the position of the control device the last time the setpoint was requested, thereby providing quick response without overshoot. Closed loop control is then used for fine tuning the process variable. The process variable is only corrected when it is measured to be outside an allowable tolerance around the setpoint. This method of closed loop control is particularly useful for situations in which the setpoint of a process variable is suddenly changed (for example an on/off scenario) and the process variable must be controlled to the new setpoint quickly and precisely.
Various embodiments of the invention also provide a method for detecting drift before it begins to affect process control. When excessive correction by a control device is required to maintain a process variable inside an allowable tolerance around a setpoint, an alarm is generated indicating that drift has occurred and control of the process variable may be lost.
Various embodiments of the invention also provide a method for minimizing the effects of hysteresis of the control device on control of the process variable. A multiplier constant is incorporated in the control logic that corrects a process variable at a different rate when ramping up to a setpoint than when ramping down to a setpoint.
Various embodiments of the invention disclose a method of providing closed loop control of a process variable while minimizing the number of calculations required to determine control device adjustment. When an adjustment of the control device is necessary, the control device is incrementally opened or closed a predetermined amount. The control device is not adjusted based on proportional, derivative, or integration calculations of the process variable and its variation from setpoint and no such calculations are performed. In addition, whenever the process variable is within an allowable tolerance around setpoint, the position of the control device is not adjusted and no calculations are required to determine the adjustment of the control device.
In one embodiment, the invention is used for closed loop control of the delivery of liquids or gases at precise flow rates, typically when the required flow is for short periods of time.
In another embodiment, the invention is used to control the flow rate of chemicals or de-ionized water (DI) to a process chamber for wet treatment of a substrate for semiconductor processing. Applications include, but are not limited to, treatment of copper-plated silicon wafers (both pre- and post-plating) and chemical/mechanical polishing of silicon wafers.
In yet another embodiment, the invention is used to control the flow rate of chemicals and DI water to an integrated bevel clean (IBC) chamber for the purpose of removing copper plated on the edge of a silicon wafer.
So that the manner in which the above recited features of the present invention can be understood in detail, a more particular description of the invention, briefly summarized above, may be had by reference to embodiments, some of which are illustrated in the appended drawings. It is to be noted, however, that the appended drawings illustrate only typical embodiments of this invention and are therefore not to be considered limiting of its scope, for the invention may admit to other equally effective embodiments.
One or more embodiments of the invention are directed to a method of closed loop control in which a process variable being controlled to a setpoint is only adjusted when the process variable's value is outside of a pre-determined deadband above or below the setpoint. Compared to PID control, various embodiments of the invention require only a small number of calculations to control the process variable to the setpoint. In one embodiment, the invention allow the process variable to be controlled to a new setpoint over a short response time without the risk of overshoot or instability. This is accomplished after a sudden change in setpoint by commanding the control device to open to a best guess position for a pre-determined length of time without using closed loop control. The best guess position is the position corresponding to the last time the process variable was controlled to the new setpoint. In another embodiment, the method further compensates for hysteresis of the control device with a single multiplication factor. In yet another embodiment, the method also detects drift of a control system prior to losing control of the process variable.
A typical closed loop control liquid delivery system applying an embodiment of the invention as its control algorithm is depicted in
The flow chart in
Referring now to
At block 706, system controller 103 reads the AI from flow meter 101. Flow meter 101 continually measures chemical flow rate and sends the latest measurement as an AI to system controller 103 periodically—in this embodiment every 100 milliseconds (block 707). System controller 103 time averages the current AI value (block 708) and calculates the current value for “flow” (block 709). In other embodiments, the value for “flow” can be time averaged over previous values instead of AI.
After “flow” is calculated, system controller 103 then checks to see if “time delay” has expired, i.e., is “time”>“time delay”? (block 710). If “time delay” has not expired, then AOnew is set to the value of AOold (block 711). The new value of AOnew is tested against the upper and lower voltage warning levels (blocks 717 and 720, respectively), AOold is updated to the latest value of AOnew (block 705), and the output signal AOnew is sent to pressure regulator 105 (block 703). This loop will continue until “time delay” has expired.
At block 710, when “time delay” expires, closed loop control of the process variable begins and the value for AOnew is determined by system controller 103 using the following logic:
After the new value for AOnew has been calculated (in block 713, 715, or 716), AOnew is tested against the upper and lower voltage warning levels (blocks 717 and 720). If AOnew<“lower voltage warning”, or if AOnew>“upper voltage warning”, then system controller 103 generates a warning indicating that the process variable is nearing the point where it can no longer be controlled (blocks 718 and 719). In other embodiments, system controller 103 generates a unique warning for “lower voltage warning” and for “upper voltage warning”.
Referring to block 705, AOold is reset to the value of AOnew, and AOnew is sent to pressure regulator 105. System controller 103 polls the condition of shut-off valve 106 again (block 701) after the required 100 millisecond interval has elapsed, and closed loop control continues.
If shut-off valve 106 is closed, system controller 103 sets the AO for the flow controller equal to 0 (full closed) and sends this command to the flow controller.
When flow is requested, shut-off valve 106 is opened and process chemical is delivered to the process chamber at a flow rate determined by pressure regulator 105. When shut-off valve 106 is opened, closed loop control is not used for a period of time equal to time delay 207. Instead, pressure regulator 105 is commanded to open to a set value for this time period and is not controlled by the system controller 103 to adjust flow closer to setpoint 201.
Time delay 207 is used because sampling rates of flow meter 101 can be much shorter than the physical response time of the flow control device (for example 100 milliseconds sampling time vs. 1 or 2 seconds control device response time). Therefore, when pressure regulator 105 is commanded to open very quickly, flow meter 101 could make a flow measurement before pressure regulator 105 has had enough time to respond to the control signal (the AO). If this happens, system controller 103 would request pressure regulator 105 to remain full open too long because of the time lag in the response of pressure regulator 105. This type of over-responsiveness of the control loop leads to overshoot past setpoint 201 whenever a large change in setpoint occurs.
Because there is no closed loop control during time delay 207, the setting for pressure regulator 105 during time delay 207 is critical. This initial setting for pressure regulator 105 (AOold) must result in flow as close as possible to setpoint 201 without the input of closed loop control. This is accomplished by setting AOold equal to the last AO sent by system controller 103 to pressure regulator 105. This “feed-forward” method of starting a process recipe step with the most recent AO value from the previous process recipe step allows pressure regulator 105 to adjust the flow as quickly as mechanically possible to a new setpoint value without the instabilities inherent in PID control. With the method of the invention, an appropriate value for time delay 207 can be selected for each situation that will minimize the potential for overshoot and begin closed loop control for fine-tuning the flow rate to the setpoint. The length of the time delay is based on the physical characteristics of the flow controller being used. Time delay 207 can be increased in length if overshoot is observed and decreased if closed loop control begins too early.
After time delay 207 has expired, system controller 103 then calculates the new AO for adjusting pressure regulator 105. Based on the latest value of flow rate 206 and setpoint 201, the new AO is determined using the control logic outlined in
System controller 103 then requests the desired flow adjustment by sending signal AOnew to voltage-to-pressure transducer 104 and flow rate 206 is adjusted by system controller 103 as described in previous paragraphs in connection with
Whenever flow rate 206 is within deadband 210 of setpoint 201, no calculations are required to determine AOnew. When flow rate 206 is outside deadband 210 of setpoint 201, the number of calculations required to determine AOnew is very low (i.e., a single addition or subtraction) compared to a typical PID calculation. This minimizes interference with the sampling time of system controller 103 due to large numbers of calculations being performed for many sensors and controllers simultaneously.
In a preferred embodiment of the invention, the invention is employed for closed loop control of chemical delivery for treatment of a substrate for semiconductor processing. In this embodiment, the substrate is a copper plated silicon wafer. The treatment is the removal of plated copper at the edge (or bevel) of the front side of the wafer and the elimination of copper sulfate and any other copper-based contamination from the back side of the edge of the wafer. The treatment typically takes place in a process chamber that is positioned on a processing platform and typically is performed immediately after the wafer has been plated. This is sometimes referred to as an IBC (integrated bevel clean) chamber. The treatment is usually defined by a process recipe that comprises all the necessary process steps for performing the treatment completely and repeatably. A typical process recipe for such a treatment comprises the following steps:
After being plated with copper in a plating cell that is also positioned on the processing platform, wafer 401 is transported by a centrally located transfer robot to the IBC chamber for treatment. Wafer 401 is placed on vacuum chuck 402 by the transfer robot and chucked by vacuum chuck 402 for the pre-rinse step. Wafer 401 is rotated on vacuum chuck 402, preferably between about 100 and about 300 rpm, and rinsed with top side and back side DI water for a short time, typically less than 15 seconds. This removes residual electrolyte that may still be on wafer 401 from the plating process. After the rinse step, wafer 401 is rotated on vacuum chuck 402 at a higher rpm, preferably between about 1000 and about 3000 rpm, for the dry step. The dry step is typically between about 5 and about 15 seconds.
The edge etch step then begins. During this step it is critical that the top side chemical flow rate is stable and very close to the desired setpoint whenever process chemical is being applied to the wafer 401. Wafer 401 is rotated at the rpm required by the process recipe, typically between about 800 and about 3000 rpm. Simultaneously, front side chemical flow begins through chemical nozzle 403 as sweep arm mechanism 406 moves nozzle arm 405 into process position 405 b at the edge of the wafer. Starting chemical flow before nozzle arm 405 is in the process position 405 b allows the chemical flow rate to stabilize to the setpoint called for by the process recipe before chemical nozzle 403 applies the process chemical to the surface of the wafer. If the top side chemical flow rate is too far below setpoint, the copper at the wafer's edge may not be completely removed. This can cause serious copper contamination problems in wafer processing platforms in which the wafer is subsequently treated. If the flow rate is too far above the set point or overshoots the setpoint, splashing may occur or the copper may be removed too far in from the edge of the wafer, either of which can destroy the electronic devices located near the edge of the wafer. If variations occur in the top side chemical flow rate during this step of the process recipe, uneven or incomplete removal of the copper often occurs. The length of time necessary to complete this step in the process recipe is dependent on several factors, including composition of top side chemical, chemical flow rate, and thickness of copper to be removed. This process recipe step typically requires between about 10 and about 60 seconds to complete the removal of copper.
For the top rinse step, top side chemical flow is stopped, top side DI flow is again initiated to remove residual top side chemical, and sweep arm mechanism 406 rotates nozzle arm 405 back to park position 405 a. Vacuum chuck 402 slows rotation to between about 100 and about 300 rpm and rinsing of the top side of the wafer continues for between about 1 and about 20 seconds.
For the back side etch step, top side DI water continues to flow and vacuum chuck 402 continues to rotate wafer 401 at the same rpm as in the top rinse step. Back side chemical is applied to the back side of the wafer for between about 5 and about 20 seconds to remove residual copper contamination. During this step, it is important that the back side chemical flow rate does not overshoot the setpoint because any back side chemical that wraps around to the front side of wafer 401 will damage electrical devices that it contacts.
For the full rinse step, top side DI water continues to flow, back side chemical is stopped, and back side DI water flow is initiated. Vacuum chuck 402 rotates wafer 401 between about 100 and about 400 rpm. This process recipe step typically lasts between about 2 and about 20 seconds.
The final dry step removes most or all moisture from the wafer. Front and back side DI water flow is stopped and wafer 401 is rotated between about 500 and about 2000 rpm. This process recipe step typically lasts between about 2 to about 20 seconds.
After treatment is complete in the IBC chamber, wafer 401 is typically transported by the centrally located transfer robot to another process for final rinsing and drying.
While the foregoing is directed to embodiments of the present invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof, and the scope thereof is determined by the claims that follow.
Citations de brevets